Add a2ui_server

[gspencergoog]

Description

This is a Genkit server that serves A2UI protocol. Now, it's not an A2A server, so that's kind of weird, but it's a start.

Summary of Changes

This pull request establishes the foundational a2ui_server package, a critical server-side component for the GenUI framework. It integrates the Genkit framework to enable dynamic UI generation by an LLM, which processes conversation history and a client's widget catalog to produce real-time UI updates. The server specifically implements the A2UI protocol, translating LLM-generated tool calls into a stream of UI modification messages for the client. This initial setup provides a robust, scalable, and well-documented backend for interactive, AI-driven user interfaces.

Highlights

• New a2ui_server Package: Introduces a new server-side component for the GenUI framework, built with Genkit.
• Dynamic UI Generation: Leverages an LLM to dynamically generate UI definitions based on conversation history and a client-provided widget catalog.
• A2UI Protocol Implementation: Serves the A2UI protocol, transforming LLM tool calls into specific UI modification messages (surfaceUpdate, beginRendering, deleteSurface).
• Streaming generateUiFlow: Implements a streaming Genkit flow (generateUiFlow) for real-time UI updates.
• Comprehensive Documentation: Includes detailed IMPLEMENTATION.md and README.md files covering architecture, API, and usage.
• Developer Tooling: Sets up ESLint for code quality, Jest for testing, and Pino for structured logging.

Changelog

• packages/spikes/a2ui_server/.gitignore
  • Added a new .gitignore file for the a2ui_server package, ignoring common build artifacts and dependencies.
• packages/spikes/a2ui_server/IMPLEMENTATION.md
  • Added a detailed implementation document outlining the purpose, architecture, schemas, API design, data flow, and logging for the a2ui_server.
• packages/spikes/a2ui_server/README.md
  • Added a comprehensive README file for the a2ui_server, including prerequisites, installation, running instructions, logging configuration, and API details.
• packages/spikes/a2ui_server/eslint.config.mjs
  • Added ESLint configuration for the TypeScript project, enforcing code style and best practices.
• packages/spikes/a2ui_server/jest.config.js
  • Added Jest configuration for unit testing the TypeScript codebase.
• packages/spikes/a2ui_server/package.json
  • Added the package manifest, defining project metadata, scripts for development and deployment, and listing all necessary dependencies and devDependencies for the Genkit server.
• packages/spikes/a2ui_server/pnpm-workspace.yaml
  • Added pnpm workspace configuration to manage dependencies within the monorepo.
• packages/spikes/a2ui_server/src/generate.ts
  • Implemented the core generateUiFlow logic, including defining UI-related tools (updateSurface, deleteSurface), dynamically generating system prompts based on the widget catalog, transforming client conversation messages into an LLM-compatible format, and streaming A2UI protocol responses.
• packages/spikes/a2ui_server/src/genkit.ts
  • Configured the Genkit environment, integrating the Google AI plugin and specifying the gemini-1.5-flash model.
• packages/spikes/a2ui_server/src/index.ts
  • Established the entry point for the Genkit flow server, starting the generateUiFlow.
• packages/spikes/a2ui_server/src/logger.ts
  • Set up a Pino-based logger for structured logging, with configurable levels and development-friendly pretty printing.
• packages/spikes/a2ui_server/src/schemas.ts
  • Defined Zod schemas for validating client requests, including a recursive JSON schema for the widget catalog and detailed schemas for various conversation message parts (text, image, UI event, UI definition).
• packages/spikes/a2ui_server/src/test/generate.test.ts
  • Added unit tests for the generateUiFlow, covering input validation, conversation mapping, tool call transformation, and response handling.
• packages/spikes/a2ui_server/test_server.sh
  • Added a utility script to facilitate testing the server's startSession endpoint with a predefined widget catalog.